xref: /freebsd/sys/netinet6/nd6.c (revision 79ac3c12a714bcd3f2354c52d948aed9575c46d6)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the project nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  *	$KAME: nd6.c,v 1.144 2001/05/24 07:44:00 itojun Exp $
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include "opt_inet.h"
38 #include "opt_inet6.h"
39 #include "opt_route.h"
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/eventhandler.h>
44 #include <sys/callout.h>
45 #include <sys/lock.h>
46 #include <sys/malloc.h>
47 #include <sys/mbuf.h>
48 #include <sys/mutex.h>
49 #include <sys/socket.h>
50 #include <sys/sockio.h>
51 #include <sys/time.h>
52 #include <sys/kernel.h>
53 #include <sys/protosw.h>
54 #include <sys/errno.h>
55 #include <sys/syslog.h>
56 #include <sys/rwlock.h>
57 #include <sys/queue.h>
58 #include <sys/sdt.h>
59 #include <sys/sysctl.h>
60 
61 #include <net/if.h>
62 #include <net/if_var.h>
63 #include <net/if_dl.h>
64 #include <net/if_types.h>
65 #include <net/route.h>
66 #include <net/route/route_ctl.h>
67 #include <net/route/nhop.h>
68 #include <net/vnet.h>
69 
70 #include <netinet/in.h>
71 #include <netinet/in_kdtrace.h>
72 #include <net/if_llatbl.h>
73 #include <netinet/if_ether.h>
74 #include <netinet6/in6_var.h>
75 #include <netinet/ip6.h>
76 #include <netinet6/ip6_var.h>
77 #include <netinet6/scope6_var.h>
78 #include <netinet6/nd6.h>
79 #include <netinet6/in6_ifattach.h>
80 #include <netinet/icmp6.h>
81 #include <netinet6/send.h>
82 
83 #include <sys/limits.h>
84 
85 #include <security/mac/mac_framework.h>
86 
87 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
88 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
89 
90 #define SIN6(s) ((const struct sockaddr_in6 *)(s))
91 
92 MALLOC_DEFINE(M_IP6NDP, "ip6ndp", "IPv6 Neighbor Discovery");
93 
94 /* timer values */
95 VNET_DEFINE(int, nd6_prune)	= 1;	/* walk list every 1 seconds */
96 VNET_DEFINE(int, nd6_delay)	= 5;	/* delay first probe time 5 second */
97 VNET_DEFINE(int, nd6_umaxtries)	= 3;	/* maximum unicast query */
98 VNET_DEFINE(int, nd6_mmaxtries)	= 3;	/* maximum multicast query */
99 VNET_DEFINE(int, nd6_useloopback) = 1;	/* use loopback interface for
100 					 * local traffic */
101 VNET_DEFINE(int, nd6_gctimer)	= (60 * 60 * 24); /* 1 day: garbage
102 					 * collection timer */
103 
104 /* preventing too many loops in ND option parsing */
105 VNET_DEFINE_STATIC(int, nd6_maxndopt) = 10; /* max # of ND options allowed */
106 
107 VNET_DEFINE(int, nd6_maxnudhint) = 0;	/* max # of subsequent upper
108 					 * layer hints */
109 VNET_DEFINE_STATIC(int, nd6_maxqueuelen) = 16; /* max pkts cached in unresolved
110 					 * ND entries */
111 #define	V_nd6_maxndopt			VNET(nd6_maxndopt)
112 #define	V_nd6_maxqueuelen		VNET(nd6_maxqueuelen)
113 
114 #ifdef ND6_DEBUG
115 VNET_DEFINE(int, nd6_debug) = 1;
116 #else
117 VNET_DEFINE(int, nd6_debug) = 0;
118 #endif
119 
120 static eventhandler_tag lle_event_eh, iflladdr_event_eh, ifnet_link_event_eh;
121 
122 VNET_DEFINE(struct nd_prhead, nd_prefix);
123 VNET_DEFINE(struct rwlock, nd6_lock);
124 VNET_DEFINE(uint64_t, nd6_list_genid);
125 VNET_DEFINE(struct mtx, nd6_onlink_mtx);
126 
127 VNET_DEFINE(int, nd6_recalc_reachtm_interval) = ND6_RECALC_REACHTM_INTERVAL;
128 #define	V_nd6_recalc_reachtm_interval	VNET(nd6_recalc_reachtm_interval)
129 
130 int	(*send_sendso_input_hook)(struct mbuf *, struct ifnet *, int, int);
131 
132 static int nd6_is_new_addr_neighbor(const struct sockaddr_in6 *,
133 	struct ifnet *);
134 static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *);
135 static void nd6_slowtimo(void *);
136 static int regen_tmpaddr(struct in6_ifaddr *);
137 static void nd6_free(struct llentry **, int);
138 static void nd6_free_redirect(const struct llentry *);
139 static void nd6_llinfo_timer(void *);
140 static void nd6_llinfo_settimer_locked(struct llentry *, long);
141 static void clear_llinfo_pqueue(struct llentry *);
142 static int nd6_resolve_slow(struct ifnet *, int, struct mbuf *,
143     const struct sockaddr_in6 *, u_char *, uint32_t *, struct llentry **);
144 static int nd6_need_cache(struct ifnet *);
145 
146 VNET_DEFINE_STATIC(struct callout, nd6_slowtimo_ch);
147 #define	V_nd6_slowtimo_ch		VNET(nd6_slowtimo_ch)
148 
149 VNET_DEFINE_STATIC(struct callout, nd6_timer_ch);
150 #define	V_nd6_timer_ch			VNET(nd6_timer_ch)
151 
152 SYSCTL_DECL(_net_inet6_icmp6);
153 
154 static void
155 nd6_lle_event(void *arg __unused, struct llentry *lle, int evt)
156 {
157 	struct rt_addrinfo rtinfo;
158 	struct sockaddr_in6 dst;
159 	struct sockaddr_dl gw;
160 	struct ifnet *ifp;
161 	int type;
162 	int fibnum;
163 
164 	LLE_WLOCK_ASSERT(lle);
165 
166 	if (lltable_get_af(lle->lle_tbl) != AF_INET6)
167 		return;
168 
169 	switch (evt) {
170 	case LLENTRY_RESOLVED:
171 		type = RTM_ADD;
172 		KASSERT(lle->la_flags & LLE_VALID,
173 		    ("%s: %p resolved but not valid?", __func__, lle));
174 		break;
175 	case LLENTRY_EXPIRED:
176 		type = RTM_DELETE;
177 		break;
178 	default:
179 		return;
180 	}
181 
182 	ifp = lltable_get_ifp(lle->lle_tbl);
183 
184 	bzero(&dst, sizeof(dst));
185 	bzero(&gw, sizeof(gw));
186 	bzero(&rtinfo, sizeof(rtinfo));
187 	lltable_fill_sa_entry(lle, (struct sockaddr *)&dst);
188 	dst.sin6_scope_id = in6_getscopezone(ifp,
189 	    in6_addrscope(&dst.sin6_addr));
190 	gw.sdl_len = sizeof(struct sockaddr_dl);
191 	gw.sdl_family = AF_LINK;
192 	gw.sdl_alen = ifp->if_addrlen;
193 	gw.sdl_index = ifp->if_index;
194 	gw.sdl_type = ifp->if_type;
195 	if (evt == LLENTRY_RESOLVED)
196 		bcopy(lle->ll_addr, gw.sdl_data, ifp->if_addrlen);
197 	rtinfo.rti_info[RTAX_DST] = (struct sockaddr *)&dst;
198 	rtinfo.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&gw;
199 	rtinfo.rti_addrs = RTA_DST | RTA_GATEWAY;
200 	fibnum = V_rt_add_addr_allfibs ? RT_ALL_FIBS : ifp->if_fib;
201 	rt_missmsg_fib(type, &rtinfo, RTF_HOST | RTF_LLDATA | (
202 	    type == RTM_ADD ? RTF_UP: 0), 0, fibnum);
203 }
204 
205 /*
206  * A handler for interface link layer address change event.
207  */
208 static void
209 nd6_iflladdr(void *arg __unused, struct ifnet *ifp)
210 {
211 	if (ifp->if_afdata[AF_INET6] == NULL)
212 		return;
213 
214 	lltable_update_ifaddr(LLTABLE6(ifp));
215 }
216 
217 void
218 nd6_init(void)
219 {
220 
221 	mtx_init(&V_nd6_onlink_mtx, "nd6 onlink", NULL, MTX_DEF);
222 	rw_init(&V_nd6_lock, "nd6 list");
223 
224 	LIST_INIT(&V_nd_prefix);
225 	nd6_defrouter_init();
226 
227 	/* Start timers. */
228 	callout_init(&V_nd6_slowtimo_ch, 0);
229 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
230 	    nd6_slowtimo, curvnet);
231 
232 	callout_init(&V_nd6_timer_ch, 0);
233 	callout_reset(&V_nd6_timer_ch, hz, nd6_timer, curvnet);
234 
235 	nd6_dad_init();
236 	if (IS_DEFAULT_VNET(curvnet)) {
237 		lle_event_eh = EVENTHANDLER_REGISTER(lle_event, nd6_lle_event,
238 		    NULL, EVENTHANDLER_PRI_ANY);
239 		iflladdr_event_eh = EVENTHANDLER_REGISTER(iflladdr_event,
240 		    nd6_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
241 		ifnet_link_event_eh = EVENTHANDLER_REGISTER(ifnet_link_event,
242 		    nd6_ifnet_link_event, NULL, EVENTHANDLER_PRI_ANY);
243 	}
244 }
245 
246 #ifdef VIMAGE
247 void
248 nd6_destroy()
249 {
250 
251 	callout_drain(&V_nd6_slowtimo_ch);
252 	callout_drain(&V_nd6_timer_ch);
253 	if (IS_DEFAULT_VNET(curvnet)) {
254 		EVENTHANDLER_DEREGISTER(ifnet_link_event, ifnet_link_event_eh);
255 		EVENTHANDLER_DEREGISTER(lle_event, lle_event_eh);
256 		EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_event_eh);
257 	}
258 	rw_destroy(&V_nd6_lock);
259 	mtx_destroy(&V_nd6_onlink_mtx);
260 }
261 #endif
262 
263 struct nd_ifinfo *
264 nd6_ifattach(struct ifnet *ifp)
265 {
266 	struct nd_ifinfo *nd;
267 
268 	nd = malloc(sizeof(*nd), M_IP6NDP, M_WAITOK | M_ZERO);
269 	nd->initialized = 1;
270 
271 	nd->chlim = IPV6_DEFHLIM;
272 	nd->basereachable = REACHABLE_TIME;
273 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
274 	nd->retrans = RETRANS_TIMER;
275 
276 	nd->flags = ND6_IFF_PERFORMNUD;
277 
278 	/* Set IPv6 disabled on all interfaces but loopback by default. */
279 	if ((ifp->if_flags & IFF_LOOPBACK) == 0)
280 		nd->flags |= ND6_IFF_IFDISABLED;
281 
282 	/* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL.
283 	 * XXXHRS: Clear ND6_IFF_AUTO_LINKLOCAL on an IFT_BRIDGE interface by
284 	 * default regardless of the V_ip6_auto_linklocal configuration to
285 	 * give a reasonable default behavior.
286 	 */
287 	if ((V_ip6_auto_linklocal && ifp->if_type != IFT_BRIDGE) ||
288 	    (ifp->if_flags & IFF_LOOPBACK))
289 		nd->flags |= ND6_IFF_AUTO_LINKLOCAL;
290 	/*
291 	 * A loopback interface does not need to accept RTADV.
292 	 * XXXHRS: Clear ND6_IFF_ACCEPT_RTADV on an IFT_BRIDGE interface by
293 	 * default regardless of the V_ip6_accept_rtadv configuration to
294 	 * prevent the interface from accepting RA messages arrived
295 	 * on one of the member interfaces with ND6_IFF_ACCEPT_RTADV.
296 	 */
297 	if (V_ip6_accept_rtadv &&
298 	    !(ifp->if_flags & IFF_LOOPBACK) &&
299 	    (ifp->if_type != IFT_BRIDGE)) {
300 			nd->flags |= ND6_IFF_ACCEPT_RTADV;
301 			/* If we globally accept rtadv, assume IPv6 on. */
302 			nd->flags &= ~ND6_IFF_IFDISABLED;
303 	}
304 	if (V_ip6_no_radr && !(ifp->if_flags & IFF_LOOPBACK))
305 		nd->flags |= ND6_IFF_NO_RADR;
306 
307 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
308 	nd6_setmtu0(ifp, nd);
309 
310 	return nd;
311 }
312 
313 void
314 nd6_ifdetach(struct ifnet *ifp, struct nd_ifinfo *nd)
315 {
316 	struct epoch_tracker et;
317 	struct ifaddr *ifa, *next;
318 
319 	NET_EPOCH_ENTER(et);
320 	CK_STAILQ_FOREACH_SAFE(ifa, &ifp->if_addrhead, ifa_link, next) {
321 		if (ifa->ifa_addr->sa_family != AF_INET6)
322 			continue;
323 
324 		/* stop DAD processing */
325 		nd6_dad_stop(ifa);
326 	}
327 	NET_EPOCH_EXIT(et);
328 
329 	free(nd, M_IP6NDP);
330 }
331 
332 /*
333  * Reset ND level link MTU. This function is called when the physical MTU
334  * changes, which means we might have to adjust the ND level MTU.
335  */
336 void
337 nd6_setmtu(struct ifnet *ifp)
338 {
339 	if (ifp->if_afdata[AF_INET6] == NULL)
340 		return;
341 
342 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
343 }
344 
345 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
346 void
347 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
348 {
349 	u_int32_t omaxmtu;
350 
351 	omaxmtu = ndi->maxmtu;
352 	ndi->maxmtu = ifp->if_mtu;
353 
354 	/*
355 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
356 	 * undesirable situation.  We thus notify the operator of the change
357 	 * explicitly.  The check for omaxmtu is necessary to restrict the
358 	 * log to the case of changing the MTU, not initializing it.
359 	 */
360 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
361 		log(LOG_NOTICE, "nd6_setmtu0: "
362 		    "new link MTU on %s (%lu) is too small for IPv6\n",
363 		    if_name(ifp), (unsigned long)ndi->maxmtu);
364 	}
365 
366 	if (ndi->maxmtu > V_in6_maxmtu)
367 		in6_setmaxmtu(); /* check all interfaces just in case */
368 
369 }
370 
371 void
372 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
373 {
374 
375 	bzero(ndopts, sizeof(*ndopts));
376 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
377 	ndopts->nd_opts_last
378 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
379 
380 	if (icmp6len == 0) {
381 		ndopts->nd_opts_done = 1;
382 		ndopts->nd_opts_search = NULL;
383 	}
384 }
385 
386 /*
387  * Take one ND option.
388  */
389 struct nd_opt_hdr *
390 nd6_option(union nd_opts *ndopts)
391 {
392 	struct nd_opt_hdr *nd_opt;
393 	int olen;
394 
395 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
396 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
397 	    __func__));
398 	if (ndopts->nd_opts_search == NULL)
399 		return NULL;
400 	if (ndopts->nd_opts_done)
401 		return NULL;
402 
403 	nd_opt = ndopts->nd_opts_search;
404 
405 	/* make sure nd_opt_len is inside the buffer */
406 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
407 		bzero(ndopts, sizeof(*ndopts));
408 		return NULL;
409 	}
410 
411 	olen = nd_opt->nd_opt_len << 3;
412 	if (olen == 0) {
413 		/*
414 		 * Message validation requires that all included
415 		 * options have a length that is greater than zero.
416 		 */
417 		bzero(ndopts, sizeof(*ndopts));
418 		return NULL;
419 	}
420 
421 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
422 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
423 		/* option overruns the end of buffer, invalid */
424 		bzero(ndopts, sizeof(*ndopts));
425 		return NULL;
426 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
427 		/* reached the end of options chain */
428 		ndopts->nd_opts_done = 1;
429 		ndopts->nd_opts_search = NULL;
430 	}
431 	return nd_opt;
432 }
433 
434 /*
435  * Parse multiple ND options.
436  * This function is much easier to use, for ND routines that do not need
437  * multiple options of the same type.
438  */
439 int
440 nd6_options(union nd_opts *ndopts)
441 {
442 	struct nd_opt_hdr *nd_opt;
443 	int i = 0;
444 
445 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
446 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
447 	    __func__));
448 	if (ndopts->nd_opts_search == NULL)
449 		return 0;
450 
451 	while (1) {
452 		nd_opt = nd6_option(ndopts);
453 		if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
454 			/*
455 			 * Message validation requires that all included
456 			 * options have a length that is greater than zero.
457 			 */
458 			ICMP6STAT_INC(icp6s_nd_badopt);
459 			bzero(ndopts, sizeof(*ndopts));
460 			return -1;
461 		}
462 
463 		if (nd_opt == NULL)
464 			goto skip1;
465 
466 		switch (nd_opt->nd_opt_type) {
467 		case ND_OPT_SOURCE_LINKADDR:
468 		case ND_OPT_TARGET_LINKADDR:
469 		case ND_OPT_MTU:
470 		case ND_OPT_REDIRECTED_HEADER:
471 		case ND_OPT_NONCE:
472 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
473 				nd6log((LOG_INFO,
474 				    "duplicated ND6 option found (type=%d)\n",
475 				    nd_opt->nd_opt_type));
476 				/* XXX bark? */
477 			} else {
478 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
479 					= nd_opt;
480 			}
481 			break;
482 		case ND_OPT_PREFIX_INFORMATION:
483 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
484 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
485 					= nd_opt;
486 			}
487 			ndopts->nd_opts_pi_end =
488 				(struct nd_opt_prefix_info *)nd_opt;
489 			break;
490 		/* What about ND_OPT_ROUTE_INFO? RFC 4191 */
491 		case ND_OPT_RDNSS:	/* RFC 6106 */
492 		case ND_OPT_DNSSL:	/* RFC 6106 */
493 			/*
494 			 * Silently ignore options we know and do not care about
495 			 * in the kernel.
496 			 */
497 			break;
498 		default:
499 			/*
500 			 * Unknown options must be silently ignored,
501 			 * to accommodate future extension to the protocol.
502 			 */
503 			nd6log((LOG_DEBUG,
504 			    "nd6_options: unsupported option %d - "
505 			    "option ignored\n", nd_opt->nd_opt_type));
506 		}
507 
508 skip1:
509 		i++;
510 		if (i > V_nd6_maxndopt) {
511 			ICMP6STAT_INC(icp6s_nd_toomanyopt);
512 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
513 			break;
514 		}
515 
516 		if (ndopts->nd_opts_done)
517 			break;
518 	}
519 
520 	return 0;
521 }
522 
523 /*
524  * ND6 timer routine to handle ND6 entries
525  */
526 static void
527 nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
528 {
529 	int canceled;
530 
531 	LLE_WLOCK_ASSERT(ln);
532 
533 	if (tick < 0) {
534 		ln->la_expire = 0;
535 		ln->ln_ntick = 0;
536 		canceled = callout_stop(&ln->lle_timer);
537 	} else {
538 		ln->la_expire = time_uptime + tick / hz;
539 		LLE_ADDREF(ln);
540 		if (tick > INT_MAX) {
541 			ln->ln_ntick = tick - INT_MAX;
542 			canceled = callout_reset(&ln->lle_timer, INT_MAX,
543 			    nd6_llinfo_timer, ln);
544 		} else {
545 			ln->ln_ntick = 0;
546 			canceled = callout_reset(&ln->lle_timer, tick,
547 			    nd6_llinfo_timer, ln);
548 		}
549 	}
550 	if (canceled > 0)
551 		LLE_REMREF(ln);
552 }
553 
554 /*
555  * Gets source address of the first packet in hold queue
556  * and stores it in @src.
557  * Returns pointer to @src (if hold queue is not empty) or NULL.
558  *
559  * Set noinline to be dtrace-friendly
560  */
561 static __noinline struct in6_addr *
562 nd6_llinfo_get_holdsrc(struct llentry *ln, struct in6_addr *src)
563 {
564 	struct ip6_hdr hdr;
565 	struct mbuf *m;
566 
567 	if (ln->la_hold == NULL)
568 		return (NULL);
569 
570 	/*
571 	 * assume every packet in la_hold has the same IP header
572 	 */
573 	m = ln->la_hold;
574 	if (sizeof(hdr) > m->m_len)
575 		return (NULL);
576 
577 	m_copydata(m, 0, sizeof(hdr), (caddr_t)&hdr);
578 	*src = hdr.ip6_src;
579 
580 	return (src);
581 }
582 
583 /*
584  * Checks if we need to switch from STALE state.
585  *
586  * RFC 4861 requires switching from STALE to DELAY state
587  * on first packet matching entry, waiting V_nd6_delay and
588  * transition to PROBE state (if upper layer confirmation was
589  * not received).
590  *
591  * This code performs a bit differently:
592  * On packet hit we don't change state (but desired state
593  * can be guessed by control plane). However, after V_nd6_delay
594  * seconds code will transition to PROBE state (so DELAY state
595  * is kinda skipped in most situations).
596  *
597  * Typically, V_nd6_gctimer is bigger than V_nd6_delay, so
598  * we perform the following upon entering STALE state:
599  *
600  * 1) Arm timer to run each V_nd6_delay seconds to make sure that
601  * if packet was transmitted at the start of given interval, we
602  * would be able to switch to PROBE state in V_nd6_delay seconds
603  * as user expects.
604  *
605  * 2) Reschedule timer until original V_nd6_gctimer expires keeping
606  * lle in STALE state (remaining timer value stored in lle_remtime).
607  *
608  * 3) Reschedule timer if packet was transmitted less that V_nd6_delay
609  * seconds ago.
610  *
611  * Returns non-zero value if the entry is still STALE (storing
612  * the next timer interval in @pdelay).
613  *
614  * Returns zero value if original timer expired or we need to switch to
615  * PROBE (store that in @do_switch variable).
616  */
617 static int
618 nd6_is_stale(struct llentry *lle, long *pdelay, int *do_switch)
619 {
620 	int nd_delay, nd_gctimer, r_skip_req;
621 	time_t lle_hittime;
622 	long delay;
623 
624 	*do_switch = 0;
625 	nd_gctimer = V_nd6_gctimer;
626 	nd_delay = V_nd6_delay;
627 
628 	LLE_REQ_LOCK(lle);
629 	r_skip_req = lle->r_skip_req;
630 	lle_hittime = lle->lle_hittime;
631 	LLE_REQ_UNLOCK(lle);
632 
633 	if (r_skip_req > 0) {
634 		/*
635 		 * Nonzero r_skip_req value was set upon entering
636 		 * STALE state. Since value was not changed, no
637 		 * packets were passed using this lle. Ask for
638 		 * timer reschedule and keep STALE state.
639 		 */
640 		delay = (long)(MIN(nd_gctimer, nd_delay));
641 		delay *= hz;
642 		if (lle->lle_remtime > delay)
643 			lle->lle_remtime -= delay;
644 		else {
645 			delay = lle->lle_remtime;
646 			lle->lle_remtime = 0;
647 		}
648 
649 		if (delay == 0) {
650 			/*
651 			 * The original ng6_gctime timeout ended,
652 			 * no more rescheduling.
653 			 */
654 			return (0);
655 		}
656 
657 		*pdelay = delay;
658 		return (1);
659 	}
660 
661 	/*
662 	 * Packet received. Verify timestamp
663 	 */
664 	delay = (long)(time_uptime - lle_hittime);
665 	if (delay < nd_delay) {
666 		/*
667 		 * V_nd6_delay still not passed since the first
668 		 * hit in STALE state.
669 		 * Reshedule timer and return.
670 		 */
671 		*pdelay = (long)(nd_delay - delay) * hz;
672 		return (1);
673 	}
674 
675 	/* Request switching to probe */
676 	*do_switch = 1;
677 	return (0);
678 }
679 
680 /*
681  * Switch @lle state to new state optionally arming timers.
682  *
683  * Set noinline to be dtrace-friendly
684  */
685 __noinline void
686 nd6_llinfo_setstate(struct llentry *lle, int newstate)
687 {
688 	struct ifnet *ifp;
689 	int nd_gctimer, nd_delay;
690 	long delay, remtime;
691 
692 	delay = 0;
693 	remtime = 0;
694 
695 	switch (newstate) {
696 	case ND6_LLINFO_INCOMPLETE:
697 		ifp = lle->lle_tbl->llt_ifp;
698 		delay = (long)ND_IFINFO(ifp)->retrans * hz / 1000;
699 		break;
700 	case ND6_LLINFO_REACHABLE:
701 		if (!ND6_LLINFO_PERMANENT(lle)) {
702 			ifp = lle->lle_tbl->llt_ifp;
703 			delay = (long)ND_IFINFO(ifp)->reachable * hz;
704 		}
705 		break;
706 	case ND6_LLINFO_STALE:
707 
708 		/*
709 		 * Notify fast path that we want to know if any packet
710 		 * is transmitted by setting r_skip_req.
711 		 */
712 		LLE_REQ_LOCK(lle);
713 		lle->r_skip_req = 1;
714 		LLE_REQ_UNLOCK(lle);
715 		nd_delay = V_nd6_delay;
716 		nd_gctimer = V_nd6_gctimer;
717 
718 		delay = (long)(MIN(nd_gctimer, nd_delay)) * hz;
719 		remtime = (long)nd_gctimer * hz - delay;
720 		break;
721 	case ND6_LLINFO_DELAY:
722 		lle->la_asked = 0;
723 		delay = (long)V_nd6_delay * hz;
724 		break;
725 	}
726 
727 	if (delay > 0)
728 		nd6_llinfo_settimer_locked(lle, delay);
729 
730 	lle->lle_remtime = remtime;
731 	lle->ln_state = newstate;
732 }
733 
734 /*
735  * Timer-dependent part of nd state machine.
736  *
737  * Set noinline to be dtrace-friendly
738  */
739 static __noinline void
740 nd6_llinfo_timer(void *arg)
741 {
742 	struct epoch_tracker et;
743 	struct llentry *ln;
744 	struct in6_addr *dst, *pdst, *psrc, src;
745 	struct ifnet *ifp;
746 	struct nd_ifinfo *ndi;
747 	int do_switch, send_ns;
748 	long delay;
749 
750 	KASSERT(arg != NULL, ("%s: arg NULL", __func__));
751 	ln = (struct llentry *)arg;
752 	ifp = lltable_get_ifp(ln->lle_tbl);
753 	CURVNET_SET(ifp->if_vnet);
754 
755 	ND6_RLOCK();
756 	LLE_WLOCK(ln);
757 	if (callout_pending(&ln->lle_timer)) {
758 		/*
759 		 * Here we are a bit odd here in the treatment of
760 		 * active/pending. If the pending bit is set, it got
761 		 * rescheduled before I ran. The active
762 		 * bit we ignore, since if it was stopped
763 		 * in ll_tablefree() and was currently running
764 		 * it would have return 0 so the code would
765 		 * not have deleted it since the callout could
766 		 * not be stopped so we want to go through
767 		 * with the delete here now. If the callout
768 		 * was restarted, the pending bit will be back on and
769 		 * we just want to bail since the callout_reset would
770 		 * return 1 and our reference would have been removed
771 		 * by nd6_llinfo_settimer_locked above since canceled
772 		 * would have been 1.
773 		 */
774 		LLE_WUNLOCK(ln);
775 		ND6_RUNLOCK();
776 		CURVNET_RESTORE();
777 		return;
778 	}
779 	NET_EPOCH_ENTER(et);
780 	ndi = ND_IFINFO(ifp);
781 	send_ns = 0;
782 	dst = &ln->r_l3addr.addr6;
783 	pdst = dst;
784 
785 	if (ln->ln_ntick > 0) {
786 		if (ln->ln_ntick > INT_MAX) {
787 			ln->ln_ntick -= INT_MAX;
788 			nd6_llinfo_settimer_locked(ln, INT_MAX);
789 		} else {
790 			ln->ln_ntick = 0;
791 			nd6_llinfo_settimer_locked(ln, ln->ln_ntick);
792 		}
793 		goto done;
794 	}
795 
796 	if (ln->la_flags & LLE_STATIC) {
797 		goto done;
798 	}
799 
800 	if (ln->la_flags & LLE_DELETED) {
801 		nd6_free(&ln, 0);
802 		goto done;
803 	}
804 
805 	switch (ln->ln_state) {
806 	case ND6_LLINFO_INCOMPLETE:
807 		if (ln->la_asked < V_nd6_mmaxtries) {
808 			ln->la_asked++;
809 			send_ns = 1;
810 			/* Send NS to multicast address */
811 			pdst = NULL;
812 		} else {
813 			struct mbuf *m = ln->la_hold;
814 			if (m) {
815 				struct mbuf *m0;
816 
817 				/*
818 				 * assuming every packet in la_hold has the
819 				 * same IP header.  Send error after unlock.
820 				 */
821 				m0 = m->m_nextpkt;
822 				m->m_nextpkt = NULL;
823 				ln->la_hold = m0;
824 				clear_llinfo_pqueue(ln);
825 			}
826 			nd6_free(&ln, 0);
827 			if (m != NULL) {
828 				struct mbuf *n = m;
829 
830 				/*
831 				 * if there are any ummapped mbufs, we
832 				 * must free them, rather than using
833 				 * them for an ICMP, as they cannot be
834 				 * checksummed.
835 				 */
836 				while ((n = n->m_next) != NULL) {
837 					if (n->m_flags & M_EXTPG)
838 						break;
839 				}
840 				if (n != NULL) {
841 					m_freem(m);
842 					m = NULL;
843 				} else {
844 					icmp6_error2(m, ICMP6_DST_UNREACH,
845 					    ICMP6_DST_UNREACH_ADDR, 0, ifp);
846 				}
847 			}
848 		}
849 		break;
850 	case ND6_LLINFO_REACHABLE:
851 		if (!ND6_LLINFO_PERMANENT(ln))
852 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
853 		break;
854 
855 	case ND6_LLINFO_STALE:
856 		if (nd6_is_stale(ln, &delay, &do_switch) != 0) {
857 			/*
858 			 * No packet has used this entry and GC timeout
859 			 * has not been passed. Reshedule timer and
860 			 * return.
861 			 */
862 			nd6_llinfo_settimer_locked(ln, delay);
863 			break;
864 		}
865 
866 		if (do_switch == 0) {
867 			/*
868 			 * GC timer has ended and entry hasn't been used.
869 			 * Run Garbage collector (RFC 4861, 5.3)
870 			 */
871 			if (!ND6_LLINFO_PERMANENT(ln))
872 				nd6_free(&ln, 1);
873 			break;
874 		}
875 
876 		/* Entry has been used AND delay timer has ended. */
877 
878 		/* FALLTHROUGH */
879 
880 	case ND6_LLINFO_DELAY:
881 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
882 			/* We need NUD */
883 			ln->la_asked = 1;
884 			nd6_llinfo_setstate(ln, ND6_LLINFO_PROBE);
885 			send_ns = 1;
886 		} else
887 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); /* XXX */
888 		break;
889 	case ND6_LLINFO_PROBE:
890 		if (ln->la_asked < V_nd6_umaxtries) {
891 			ln->la_asked++;
892 			send_ns = 1;
893 		} else {
894 			nd6_free(&ln, 0);
895 		}
896 		break;
897 	default:
898 		panic("%s: paths in a dark night can be confusing: %d",
899 		    __func__, ln->ln_state);
900 	}
901 done:
902 	if (ln != NULL)
903 		ND6_RUNLOCK();
904 	if (send_ns != 0) {
905 		nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
906 		psrc = nd6_llinfo_get_holdsrc(ln, &src);
907 		LLE_FREE_LOCKED(ln);
908 		ln = NULL;
909 		nd6_ns_output(ifp, psrc, pdst, dst, NULL);
910 	}
911 
912 	if (ln != NULL)
913 		LLE_FREE_LOCKED(ln);
914 	NET_EPOCH_EXIT(et);
915 	CURVNET_RESTORE();
916 }
917 
918 /*
919  * ND6 timer routine to expire default route list and prefix list
920  */
921 void
922 nd6_timer(void *arg)
923 {
924 	CURVNET_SET((struct vnet *) arg);
925 	struct epoch_tracker et;
926 	struct nd_prhead prl;
927 	struct nd_prefix *pr, *npr;
928 	struct ifnet *ifp;
929 	struct in6_ifaddr *ia6, *nia6;
930 	uint64_t genid;
931 
932 	LIST_INIT(&prl);
933 
934 	NET_EPOCH_ENTER(et);
935 	nd6_defrouter_timer();
936 
937 	/*
938 	 * expire interface addresses.
939 	 * in the past the loop was inside prefix expiry processing.
940 	 * However, from a stricter speci-confrmance standpoint, we should
941 	 * rather separate address lifetimes and prefix lifetimes.
942 	 *
943 	 * XXXRW: in6_ifaddrhead locking.
944 	 */
945   addrloop:
946 	CK_STAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) {
947 		/* check address lifetime */
948 		if (IFA6_IS_INVALID(ia6)) {
949 			int regen = 0;
950 
951 			/*
952 			 * If the expiring address is temporary, try
953 			 * regenerating a new one.  This would be useful when
954 			 * we suspended a laptop PC, then turned it on after a
955 			 * period that could invalidate all temporary
956 			 * addresses.  Although we may have to restart the
957 			 * loop (see below), it must be after purging the
958 			 * address.  Otherwise, we'd see an infinite loop of
959 			 * regeneration.
960 			 */
961 			if (V_ip6_use_tempaddr &&
962 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
963 				if (regen_tmpaddr(ia6) == 0)
964 					regen = 1;
965 			}
966 
967 			in6_purgeaddr(&ia6->ia_ifa);
968 
969 			if (regen)
970 				goto addrloop; /* XXX: see below */
971 		} else if (IFA6_IS_DEPRECATED(ia6)) {
972 			int oldflags = ia6->ia6_flags;
973 
974 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
975 
976 			/*
977 			 * If a temporary address has just become deprecated,
978 			 * regenerate a new one if possible.
979 			 */
980 			if (V_ip6_use_tempaddr &&
981 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
982 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
983 				if (regen_tmpaddr(ia6) == 0) {
984 					/*
985 					 * A new temporary address is
986 					 * generated.
987 					 * XXX: this means the address chain
988 					 * has changed while we are still in
989 					 * the loop.  Although the change
990 					 * would not cause disaster (because
991 					 * it's not a deletion, but an
992 					 * addition,) we'd rather restart the
993 					 * loop just for safety.  Or does this
994 					 * significantly reduce performance??
995 					 */
996 					goto addrloop;
997 				}
998 			}
999 		} else if ((ia6->ia6_flags & IN6_IFF_TENTATIVE) != 0) {
1000 			/*
1001 			 * Schedule DAD for a tentative address.  This happens
1002 			 * if the interface was down or not running
1003 			 * when the address was configured.
1004 			 */
1005 			int delay;
1006 
1007 			delay = arc4random() %
1008 			    (MAX_RTR_SOLICITATION_DELAY * hz);
1009 			nd6_dad_start((struct ifaddr *)ia6, delay);
1010 		} else {
1011 			/*
1012 			 * Check status of the interface.  If it is down,
1013 			 * mark the address as tentative for future DAD.
1014 			 */
1015 			ifp = ia6->ia_ifp;
1016 			if ((ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0 &&
1017 			    ((ifp->if_flags & IFF_UP) == 0 ||
1018 			    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1019 			    (ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) != 0)){
1020 				ia6->ia6_flags &= ~IN6_IFF_DUPLICATED;
1021 				ia6->ia6_flags |= IN6_IFF_TENTATIVE;
1022 			}
1023 
1024 			/*
1025 			 * A new RA might have made a deprecated address
1026 			 * preferred.
1027 			 */
1028 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
1029 		}
1030 	}
1031 	NET_EPOCH_EXIT(et);
1032 
1033 	ND6_WLOCK();
1034 restart:
1035 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1036 		/*
1037 		 * Expire prefixes. Since the pltime is only used for
1038 		 * autoconfigured addresses, pltime processing for prefixes is
1039 		 * not necessary.
1040 		 *
1041 		 * Only unlink after all derived addresses have expired. This
1042 		 * may not occur until two hours after the prefix has expired
1043 		 * per RFC 4862. If the prefix expires before its derived
1044 		 * addresses, mark it off-link. This will be done automatically
1045 		 * after unlinking if no address references remain.
1046 		 */
1047 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME ||
1048 		    time_uptime - pr->ndpr_lastupdate <= pr->ndpr_vltime)
1049 			continue;
1050 
1051 		if (pr->ndpr_addrcnt == 0) {
1052 			nd6_prefix_unlink(pr, &prl);
1053 			continue;
1054 		}
1055 		if ((pr->ndpr_stateflags & NDPRF_ONLINK) != 0) {
1056 			genid = V_nd6_list_genid;
1057 			nd6_prefix_ref(pr);
1058 			ND6_WUNLOCK();
1059 			ND6_ONLINK_LOCK();
1060 			(void)nd6_prefix_offlink(pr);
1061 			ND6_ONLINK_UNLOCK();
1062 			ND6_WLOCK();
1063 			nd6_prefix_rele(pr);
1064 			if (genid != V_nd6_list_genid)
1065 				goto restart;
1066 		}
1067 	}
1068 	ND6_WUNLOCK();
1069 
1070 	while ((pr = LIST_FIRST(&prl)) != NULL) {
1071 		LIST_REMOVE(pr, ndpr_entry);
1072 		nd6_prefix_del(pr);
1073 	}
1074 
1075 	callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
1076 	    nd6_timer, curvnet);
1077 
1078 	CURVNET_RESTORE();
1079 }
1080 
1081 /*
1082  * ia6 - deprecated/invalidated temporary address
1083  */
1084 static int
1085 regen_tmpaddr(struct in6_ifaddr *ia6)
1086 {
1087 	struct ifaddr *ifa;
1088 	struct ifnet *ifp;
1089 	struct in6_ifaddr *public_ifa6 = NULL;
1090 
1091 	NET_EPOCH_ASSERT();
1092 
1093 	ifp = ia6->ia_ifa.ifa_ifp;
1094 	CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1095 		struct in6_ifaddr *it6;
1096 
1097 		if (ifa->ifa_addr->sa_family != AF_INET6)
1098 			continue;
1099 
1100 		it6 = (struct in6_ifaddr *)ifa;
1101 
1102 		/* ignore no autoconf addresses. */
1103 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1104 			continue;
1105 
1106 		/* ignore autoconf addresses with different prefixes. */
1107 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
1108 			continue;
1109 
1110 		/*
1111 		 * Now we are looking at an autoconf address with the same
1112 		 * prefix as ours.  If the address is temporary and is still
1113 		 * preferred, do not create another one.  It would be rare, but
1114 		 * could happen, for example, when we resume a laptop PC after
1115 		 * a long period.
1116 		 */
1117 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
1118 		    !IFA6_IS_DEPRECATED(it6)) {
1119 			public_ifa6 = NULL;
1120 			break;
1121 		}
1122 
1123 		/*
1124 		 * This is a public autoconf address that has the same prefix
1125 		 * as ours.  If it is preferred, keep it.  We can't break the
1126 		 * loop here, because there may be a still-preferred temporary
1127 		 * address with the prefix.
1128 		 */
1129 		if (!IFA6_IS_DEPRECATED(it6))
1130 			public_ifa6 = it6;
1131 	}
1132 	if (public_ifa6 != NULL)
1133 		ifa_ref(&public_ifa6->ia_ifa);
1134 
1135 	if (public_ifa6 != NULL) {
1136 		int e;
1137 
1138 		if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
1139 			ifa_free(&public_ifa6->ia_ifa);
1140 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
1141 			    " tmp addr,errno=%d\n", e);
1142 			return (-1);
1143 		}
1144 		ifa_free(&public_ifa6->ia_ifa);
1145 		return (0);
1146 	}
1147 
1148 	return (-1);
1149 }
1150 
1151 /*
1152  * Remove prefix and default router list entries corresponding to ifp. Neighbor
1153  * cache entries are freed in in6_domifdetach().
1154  */
1155 void
1156 nd6_purge(struct ifnet *ifp)
1157 {
1158 	struct nd_prhead prl;
1159 	struct nd_prefix *pr, *npr;
1160 
1161 	LIST_INIT(&prl);
1162 
1163 	/* Purge default router list entries toward ifp. */
1164 	nd6_defrouter_purge(ifp);
1165 
1166 	ND6_WLOCK();
1167 	/*
1168 	 * Remove prefixes on ifp. We should have already removed addresses on
1169 	 * this interface, so no addresses should be referencing these prefixes.
1170 	 */
1171 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1172 		if (pr->ndpr_ifp == ifp)
1173 			nd6_prefix_unlink(pr, &prl);
1174 	}
1175 	ND6_WUNLOCK();
1176 
1177 	/* Delete the unlinked prefix objects. */
1178 	while ((pr = LIST_FIRST(&prl)) != NULL) {
1179 		LIST_REMOVE(pr, ndpr_entry);
1180 		nd6_prefix_del(pr);
1181 	}
1182 
1183 	/* cancel default outgoing interface setting */
1184 	if (V_nd6_defifindex == ifp->if_index)
1185 		nd6_setdefaultiface(0);
1186 
1187 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1188 		/* Refresh default router list. */
1189 		defrouter_select_fib(ifp->if_fib);
1190 	}
1191 }
1192 
1193 /*
1194  * the caller acquires and releases the lock on the lltbls
1195  * Returns the llentry locked
1196  */
1197 struct llentry *
1198 nd6_lookup(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1199 {
1200 	struct sockaddr_in6 sin6;
1201 	struct llentry *ln;
1202 
1203 	bzero(&sin6, sizeof(sin6));
1204 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1205 	sin6.sin6_family = AF_INET6;
1206 	sin6.sin6_addr = *addr6;
1207 
1208 	IF_AFDATA_LOCK_ASSERT(ifp);
1209 
1210 	ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)&sin6);
1211 
1212 	return (ln);
1213 }
1214 
1215 static struct llentry *
1216 nd6_alloc(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1217 {
1218 	struct sockaddr_in6 sin6;
1219 	struct llentry *ln;
1220 
1221 	bzero(&sin6, sizeof(sin6));
1222 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1223 	sin6.sin6_family = AF_INET6;
1224 	sin6.sin6_addr = *addr6;
1225 
1226 	ln = lltable_alloc_entry(LLTABLE6(ifp), 0, (struct sockaddr *)&sin6);
1227 	if (ln != NULL)
1228 		ln->ln_state = ND6_LLINFO_NOSTATE;
1229 
1230 	return (ln);
1231 }
1232 
1233 /*
1234  * Test whether a given IPv6 address is a neighbor or not, ignoring
1235  * the actual neighbor cache.  The neighbor cache is ignored in order
1236  * to not reenter the routing code from within itself.
1237  */
1238 static int
1239 nd6_is_new_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1240 {
1241 	struct nd_prefix *pr;
1242 	struct ifaddr *ifa;
1243 	struct rt_addrinfo info;
1244 	struct sockaddr_in6 rt_key;
1245 	const struct sockaddr *dst6;
1246 	uint64_t genid;
1247 	int error, fibnum;
1248 
1249 	/*
1250 	 * A link-local address is always a neighbor.
1251 	 * XXX: a link does not necessarily specify a single interface.
1252 	 */
1253 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
1254 		struct sockaddr_in6 sin6_copy;
1255 		u_int32_t zone;
1256 
1257 		/*
1258 		 * We need sin6_copy since sa6_recoverscope() may modify the
1259 		 * content (XXX).
1260 		 */
1261 		sin6_copy = *addr;
1262 		if (sa6_recoverscope(&sin6_copy))
1263 			return (0); /* XXX: should be impossible */
1264 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
1265 			return (0);
1266 		if (sin6_copy.sin6_scope_id == zone)
1267 			return (1);
1268 		else
1269 			return (0);
1270 	}
1271 
1272 	bzero(&rt_key, sizeof(rt_key));
1273 	bzero(&info, sizeof(info));
1274 	info.rti_info[RTAX_DST] = (struct sockaddr *)&rt_key;
1275 
1276 	/*
1277 	 * If the address matches one of our addresses,
1278 	 * it should be a neighbor.
1279 	 * If the address matches one of our on-link prefixes, it should be a
1280 	 * neighbor.
1281 	 */
1282 	ND6_RLOCK();
1283 restart:
1284 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
1285 		if (pr->ndpr_ifp != ifp)
1286 			continue;
1287 
1288 		if ((pr->ndpr_stateflags & NDPRF_ONLINK) == 0) {
1289 			dst6 = (const struct sockaddr *)&pr->ndpr_prefix;
1290 
1291 			/*
1292 			 * We only need to check all FIBs if add_addr_allfibs
1293 			 * is unset. If set, checking any FIB will suffice.
1294 			 */
1295 			fibnum = V_rt_add_addr_allfibs ? rt_numfibs - 1 : 0;
1296 			for (; fibnum < rt_numfibs; fibnum++) {
1297 				genid = V_nd6_list_genid;
1298 				ND6_RUNLOCK();
1299 
1300 				/*
1301 				 * Restore length field before
1302 				 * retrying lookup
1303 				 */
1304 				rt_key.sin6_len = sizeof(rt_key);
1305 				error = rib_lookup_info(fibnum, dst6, 0, 0,
1306 						        &info);
1307 
1308 				ND6_RLOCK();
1309 				if (genid != V_nd6_list_genid)
1310 					goto restart;
1311 				if (error == 0)
1312 					break;
1313 			}
1314 			if (error != 0)
1315 				continue;
1316 
1317 			/*
1318 			 * This is the case where multiple interfaces
1319 			 * have the same prefix, but only one is installed
1320 			 * into the routing table and that prefix entry
1321 			 * is not the one being examined here.
1322 			 */
1323 			if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1324 			    &rt_key.sin6_addr))
1325 				continue;
1326 		}
1327 
1328 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1329 		    &addr->sin6_addr, &pr->ndpr_mask)) {
1330 			ND6_RUNLOCK();
1331 			return (1);
1332 		}
1333 	}
1334 	ND6_RUNLOCK();
1335 
1336 	/*
1337 	 * If the address is assigned on the node of the other side of
1338 	 * a p2p interface, the address should be a neighbor.
1339 	 */
1340 	if (ifp->if_flags & IFF_POINTOPOINT) {
1341 		struct epoch_tracker et;
1342 
1343 		NET_EPOCH_ENTER(et);
1344 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1345 			if (ifa->ifa_addr->sa_family != addr->sin6_family)
1346 				continue;
1347 			if (ifa->ifa_dstaddr != NULL &&
1348 			    sa_equal(addr, ifa->ifa_dstaddr)) {
1349 				NET_EPOCH_EXIT(et);
1350 				return 1;
1351 			}
1352 		}
1353 		NET_EPOCH_EXIT(et);
1354 	}
1355 
1356 	/*
1357 	 * If the default router list is empty, all addresses are regarded
1358 	 * as on-link, and thus, as a neighbor.
1359 	 */
1360 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV &&
1361 	    nd6_defrouter_list_empty() &&
1362 	    V_nd6_defifindex == ifp->if_index) {
1363 		return (1);
1364 	}
1365 
1366 	return (0);
1367 }
1368 
1369 /*
1370  * Detect if a given IPv6 address identifies a neighbor on a given link.
1371  * XXX: should take care of the destination of a p2p link?
1372  */
1373 int
1374 nd6_is_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1375 {
1376 	struct llentry *lle;
1377 	int rc = 0;
1378 
1379 	NET_EPOCH_ASSERT();
1380 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1381 	if (nd6_is_new_addr_neighbor(addr, ifp))
1382 		return (1);
1383 
1384 	/*
1385 	 * Even if the address matches none of our addresses, it might be
1386 	 * in the neighbor cache.
1387 	 */
1388 	if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) {
1389 		LLE_RUNLOCK(lle);
1390 		rc = 1;
1391 	}
1392 	return (rc);
1393 }
1394 
1395 /*
1396  * Free an nd6 llinfo entry.
1397  * Since the function would cause significant changes in the kernel, DO NOT
1398  * make it global, unless you have a strong reason for the change, and are sure
1399  * that the change is safe.
1400  *
1401  * Set noinline to be dtrace-friendly
1402  */
1403 static __noinline void
1404 nd6_free(struct llentry **lnp, int gc)
1405 {
1406 	struct ifnet *ifp;
1407 	struct llentry *ln;
1408 	struct nd_defrouter *dr;
1409 
1410 	ln = *lnp;
1411 	*lnp = NULL;
1412 
1413 	LLE_WLOCK_ASSERT(ln);
1414 	ND6_RLOCK_ASSERT();
1415 
1416 	ifp = lltable_get_ifp(ln->lle_tbl);
1417 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) != 0)
1418 		dr = defrouter_lookup_locked(&ln->r_l3addr.addr6, ifp);
1419 	else
1420 		dr = NULL;
1421 	ND6_RUNLOCK();
1422 
1423 	if ((ln->la_flags & LLE_DELETED) == 0)
1424 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
1425 
1426 	/*
1427 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1428 	 * even though it is not harmful, it was not really necessary.
1429 	 */
1430 
1431 	/* cancel timer */
1432 	nd6_llinfo_settimer_locked(ln, -1);
1433 
1434 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1435 		if (dr != NULL && dr->expire &&
1436 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
1437 			/*
1438 			 * If the reason for the deletion is just garbage
1439 			 * collection, and the neighbor is an active default
1440 			 * router, do not delete it.  Instead, reset the GC
1441 			 * timer using the router's lifetime.
1442 			 * Simply deleting the entry would affect default
1443 			 * router selection, which is not necessarily a good
1444 			 * thing, especially when we're using router preference
1445 			 * values.
1446 			 * XXX: the check for ln_state would be redundant,
1447 			 *      but we intentionally keep it just in case.
1448 			 */
1449 			if (dr->expire > time_uptime)
1450 				nd6_llinfo_settimer_locked(ln,
1451 				    (dr->expire - time_uptime) * hz);
1452 			else
1453 				nd6_llinfo_settimer_locked(ln,
1454 				    (long)V_nd6_gctimer * hz);
1455 
1456 			LLE_REMREF(ln);
1457 			LLE_WUNLOCK(ln);
1458 			defrouter_rele(dr);
1459 			return;
1460 		}
1461 
1462 		if (dr) {
1463 			/*
1464 			 * Unreachablity of a router might affect the default
1465 			 * router selection and on-link detection of advertised
1466 			 * prefixes.
1467 			 */
1468 
1469 			/*
1470 			 * Temporarily fake the state to choose a new default
1471 			 * router and to perform on-link determination of
1472 			 * prefixes correctly.
1473 			 * Below the state will be set correctly,
1474 			 * or the entry itself will be deleted.
1475 			 */
1476 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1477 		}
1478 
1479 		if (ln->ln_router || dr) {
1480 			/*
1481 			 * We need to unlock to avoid a LOR with rt6_flush() with the
1482 			 * rnh and for the calls to pfxlist_onlink_check() and
1483 			 * defrouter_select_fib() in the block further down for calls
1484 			 * into nd6_lookup().  We still hold a ref.
1485 			 */
1486 			LLE_WUNLOCK(ln);
1487 
1488 			/*
1489 			 * rt6_flush must be called whether or not the neighbor
1490 			 * is in the Default Router List.
1491 			 * See a corresponding comment in nd6_na_input().
1492 			 */
1493 			rt6_flush(&ln->r_l3addr.addr6, ifp);
1494 		}
1495 
1496 		if (dr) {
1497 			/*
1498 			 * Since defrouter_select_fib() does not affect the
1499 			 * on-link determination and MIP6 needs the check
1500 			 * before the default router selection, we perform
1501 			 * the check now.
1502 			 */
1503 			pfxlist_onlink_check();
1504 
1505 			/*
1506 			 * Refresh default router list.
1507 			 */
1508 			defrouter_select_fib(dr->ifp->if_fib);
1509 		}
1510 
1511 		/*
1512 		 * If this entry was added by an on-link redirect, remove the
1513 		 * corresponding host route.
1514 		 */
1515 		if (ln->la_flags & LLE_REDIRECT)
1516 			nd6_free_redirect(ln);
1517 
1518 		if (ln->ln_router || dr)
1519 			LLE_WLOCK(ln);
1520 	}
1521 
1522 	/*
1523 	 * Save to unlock. We still hold an extra reference and will not
1524 	 * free(9) in llentry_free() if someone else holds one as well.
1525 	 */
1526 	LLE_WUNLOCK(ln);
1527 	IF_AFDATA_LOCK(ifp);
1528 	LLE_WLOCK(ln);
1529 	/* Guard against race with other llentry_free(). */
1530 	if (ln->la_flags & LLE_LINKED) {
1531 		/* Remove callout reference */
1532 		LLE_REMREF(ln);
1533 		lltable_unlink_entry(ln->lle_tbl, ln);
1534 	}
1535 	IF_AFDATA_UNLOCK(ifp);
1536 
1537 	llentry_free(ln);
1538 	if (dr != NULL)
1539 		defrouter_rele(dr);
1540 }
1541 
1542 static int
1543 nd6_isdynrte(const struct rtentry *rt, const struct nhop_object *nh, void *xap)
1544 {
1545 
1546 	if (nh->nh_flags & NHF_REDIRECT)
1547 		return (1);
1548 
1549 	return (0);
1550 }
1551 
1552 /*
1553  * Remove the rtentry for the given llentry,
1554  * both of which were installed by a redirect.
1555  */
1556 static void
1557 nd6_free_redirect(const struct llentry *ln)
1558 {
1559 	int fibnum;
1560 	struct sockaddr_in6 sin6;
1561 	struct rt_addrinfo info;
1562 	struct rib_cmd_info rc;
1563 	struct epoch_tracker et;
1564 
1565 	lltable_fill_sa_entry(ln, (struct sockaddr *)&sin6);
1566 	memset(&info, 0, sizeof(info));
1567 	info.rti_info[RTAX_DST] = (struct sockaddr *)&sin6;
1568 	info.rti_filter = nd6_isdynrte;
1569 
1570 	NET_EPOCH_ENTER(et);
1571 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++)
1572 		rib_action(fibnum, RTM_DELETE, &info, &rc);
1573 	NET_EPOCH_EXIT(et);
1574 }
1575 
1576 /*
1577  * Updates status of the default router route.
1578  */
1579 static void
1580 check_release_defrouter(struct rib_cmd_info *rc, void *_cbdata)
1581 {
1582 	struct nd_defrouter *dr;
1583 	struct nhop_object *nh;
1584 
1585 	nh = rc->rc_nh_old;
1586 
1587 	if ((nh != NULL) && (nh->nh_flags & NHF_DEFAULT)) {
1588 		dr = defrouter_lookup(&nh->gw6_sa.sin6_addr, nh->nh_ifp);
1589 		if (dr != NULL) {
1590 			dr->installed = 0;
1591 			defrouter_rele(dr);
1592 		}
1593 	}
1594 }
1595 
1596 void
1597 nd6_subscription_cb(struct rib_head *rnh, struct rib_cmd_info *rc, void *arg)
1598 {
1599 
1600 #ifdef ROUTE_MPATH
1601 	rib_decompose_notification(rc, check_release_defrouter, NULL);
1602 #else
1603 	check_release_defrouter(rc, NULL);
1604 #endif
1605 }
1606 
1607 int
1608 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1609 {
1610 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1611 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1612 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1613 	struct epoch_tracker et;
1614 	int error = 0;
1615 
1616 	if (ifp->if_afdata[AF_INET6] == NULL)
1617 		return (EPFNOSUPPORT);
1618 	switch (cmd) {
1619 	case OSIOCGIFINFO_IN6:
1620 #define ND	ndi->ndi
1621 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1622 		bzero(&ND, sizeof(ND));
1623 		ND.linkmtu = IN6_LINKMTU(ifp);
1624 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1625 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1626 		ND.reachable = ND_IFINFO(ifp)->reachable;
1627 		ND.retrans = ND_IFINFO(ifp)->retrans;
1628 		ND.flags = ND_IFINFO(ifp)->flags;
1629 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1630 		ND.chlim = ND_IFINFO(ifp)->chlim;
1631 		break;
1632 	case SIOCGIFINFO_IN6:
1633 		ND = *ND_IFINFO(ifp);
1634 		break;
1635 	case SIOCSIFINFO_IN6:
1636 		/*
1637 		 * used to change host variables from userland.
1638 		 * intended for a use on router to reflect RA configurations.
1639 		 */
1640 		/* 0 means 'unspecified' */
1641 		if (ND.linkmtu != 0) {
1642 			if (ND.linkmtu < IPV6_MMTU ||
1643 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1644 				error = EINVAL;
1645 				break;
1646 			}
1647 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1648 		}
1649 
1650 		if (ND.basereachable != 0) {
1651 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1652 
1653 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1654 			if (ND.basereachable != obasereachable)
1655 				ND_IFINFO(ifp)->reachable =
1656 				    ND_COMPUTE_RTIME(ND.basereachable);
1657 		}
1658 		if (ND.retrans != 0)
1659 			ND_IFINFO(ifp)->retrans = ND.retrans;
1660 		if (ND.chlim != 0)
1661 			ND_IFINFO(ifp)->chlim = ND.chlim;
1662 		/* FALLTHROUGH */
1663 	case SIOCSIFINFO_FLAGS:
1664 	{
1665 		struct ifaddr *ifa;
1666 		struct in6_ifaddr *ia;
1667 
1668 		if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1669 		    !(ND.flags & ND6_IFF_IFDISABLED)) {
1670 			/* ifdisabled 1->0 transision */
1671 
1672 			/*
1673 			 * If the interface is marked as ND6_IFF_IFDISABLED and
1674 			 * has an link-local address with IN6_IFF_DUPLICATED,
1675 			 * do not clear ND6_IFF_IFDISABLED.
1676 			 * See RFC 4862, Section 5.4.5.
1677 			 */
1678 			NET_EPOCH_ENTER(et);
1679 			CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1680 				if (ifa->ifa_addr->sa_family != AF_INET6)
1681 					continue;
1682 				ia = (struct in6_ifaddr *)ifa;
1683 				if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1684 				    IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1685 					break;
1686 			}
1687 			NET_EPOCH_EXIT(et);
1688 
1689 			if (ifa != NULL) {
1690 				/* LLA is duplicated. */
1691 				ND.flags |= ND6_IFF_IFDISABLED;
1692 				log(LOG_ERR, "Cannot enable an interface"
1693 				    " with a link-local address marked"
1694 				    " duplicate.\n");
1695 			} else {
1696 				ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1697 				if (ifp->if_flags & IFF_UP)
1698 					in6_if_up(ifp);
1699 			}
1700 		} else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1701 			    (ND.flags & ND6_IFF_IFDISABLED)) {
1702 			/* ifdisabled 0->1 transision */
1703 			/* Mark all IPv6 address as tentative. */
1704 
1705 			ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1706 			if (V_ip6_dad_count > 0 &&
1707 			    (ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0) {
1708 				NET_EPOCH_ENTER(et);
1709 				CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead,
1710 				    ifa_link) {
1711 					if (ifa->ifa_addr->sa_family !=
1712 					    AF_INET6)
1713 						continue;
1714 					ia = (struct in6_ifaddr *)ifa;
1715 					ia->ia6_flags |= IN6_IFF_TENTATIVE;
1716 				}
1717 				NET_EPOCH_EXIT(et);
1718 			}
1719 		}
1720 
1721 		if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1722 			if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1723 				/* auto_linklocal 0->1 transision */
1724 
1725 				/* If no link-local address on ifp, configure */
1726 				ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1727 				in6_ifattach(ifp, NULL);
1728 			} else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1729 			    ifp->if_flags & IFF_UP) {
1730 				/*
1731 				 * When the IF already has
1732 				 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1733 				 * address is assigned, and IFF_UP, try to
1734 				 * assign one.
1735 				 */
1736 				NET_EPOCH_ENTER(et);
1737 				CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead,
1738 				    ifa_link) {
1739 					if (ifa->ifa_addr->sa_family !=
1740 					    AF_INET6)
1741 						continue;
1742 					ia = (struct in6_ifaddr *)ifa;
1743 					if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1744 						break;
1745 				}
1746 				NET_EPOCH_EXIT(et);
1747 				if (ifa != NULL)
1748 					/* No LLA is configured. */
1749 					in6_ifattach(ifp, NULL);
1750 			}
1751 		}
1752 		ND_IFINFO(ifp)->flags = ND.flags;
1753 		break;
1754 	}
1755 #undef ND
1756 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1757 		/* sync kernel routing table with the default router list */
1758 		defrouter_reset();
1759 		defrouter_select_fib(RT_ALL_FIBS);
1760 		break;
1761 	case SIOCSPFXFLUSH_IN6:
1762 	{
1763 		/* flush all the prefix advertised by routers */
1764 		struct in6_ifaddr *ia, *ia_next;
1765 		struct nd_prefix *pr, *next;
1766 		struct nd_prhead prl;
1767 
1768 		LIST_INIT(&prl);
1769 
1770 		ND6_WLOCK();
1771 		LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) {
1772 			if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1773 				continue; /* XXX */
1774 			nd6_prefix_unlink(pr, &prl);
1775 		}
1776 		ND6_WUNLOCK();
1777 
1778 		while ((pr = LIST_FIRST(&prl)) != NULL) {
1779 			LIST_REMOVE(pr, ndpr_entry);
1780 			/* XXXRW: in6_ifaddrhead locking. */
1781 			CK_STAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1782 			    ia_next) {
1783 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1784 					continue;
1785 
1786 				if (ia->ia6_ndpr == pr)
1787 					in6_purgeaddr(&ia->ia_ifa);
1788 			}
1789 			nd6_prefix_del(pr);
1790 		}
1791 		break;
1792 	}
1793 	case SIOCSRTRFLUSH_IN6:
1794 	{
1795 		/* flush all the default routers */
1796 
1797 		defrouter_reset();
1798 		nd6_defrouter_flush_all();
1799 		defrouter_select_fib(RT_ALL_FIBS);
1800 		break;
1801 	}
1802 	case SIOCGNBRINFO_IN6:
1803 	{
1804 		struct llentry *ln;
1805 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1806 
1807 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1808 			return (error);
1809 
1810 		NET_EPOCH_ENTER(et);
1811 		ln = nd6_lookup(&nb_addr, 0, ifp);
1812 		NET_EPOCH_EXIT(et);
1813 
1814 		if (ln == NULL) {
1815 			error = EINVAL;
1816 			break;
1817 		}
1818 		nbi->state = ln->ln_state;
1819 		nbi->asked = ln->la_asked;
1820 		nbi->isrouter = ln->ln_router;
1821 		if (ln->la_expire == 0)
1822 			nbi->expire = 0;
1823 		else
1824 			nbi->expire = ln->la_expire + ln->lle_remtime / hz +
1825 			    (time_second - time_uptime);
1826 		LLE_RUNLOCK(ln);
1827 		break;
1828 	}
1829 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1830 		ndif->ifindex = V_nd6_defifindex;
1831 		break;
1832 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1833 		return (nd6_setdefaultiface(ndif->ifindex));
1834 	}
1835 	return (error);
1836 }
1837 
1838 /*
1839  * Calculates new isRouter value based on provided parameters and
1840  * returns it.
1841  */
1842 static int
1843 nd6_is_router(int type, int code, int is_new, int old_addr, int new_addr,
1844     int ln_router)
1845 {
1846 
1847 	/*
1848 	 * ICMP6 type dependent behavior.
1849 	 *
1850 	 * NS: clear IsRouter if new entry
1851 	 * RS: clear IsRouter
1852 	 * RA: set IsRouter if there's lladdr
1853 	 * redir: clear IsRouter if new entry
1854 	 *
1855 	 * RA case, (1):
1856 	 * The spec says that we must set IsRouter in the following cases:
1857 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1858 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1859 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1860 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1861 	 * neighbor cache, this is similar to (6).
1862 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1863 	 *
1864 	 *   is_new  old_addr new_addr 	    NS  RS  RA	redir
1865 	 *							D R
1866 	 *	0	n	n	(1)	c   ?     s
1867 	 *	0	y	n	(2)	c   s     s
1868 	 *	0	n	y	(3)	c   s     s
1869 	 *	0	y	y	(4)	c   s     s
1870 	 *	0	y	y	(5)	c   s     s
1871 	 *	1	--	n	(6) c	c	c s
1872 	 *	1	--	y	(7) c	c   s	c s
1873 	 *
1874 	 *					(c=clear s=set)
1875 	 */
1876 	switch (type & 0xff) {
1877 	case ND_NEIGHBOR_SOLICIT:
1878 		/*
1879 		 * New entry must have is_router flag cleared.
1880 		 */
1881 		if (is_new)					/* (6-7) */
1882 			ln_router = 0;
1883 		break;
1884 	case ND_REDIRECT:
1885 		/*
1886 		 * If the icmp is a redirect to a better router, always set the
1887 		 * is_router flag.  Otherwise, if the entry is newly created,
1888 		 * clear the flag.  [RFC 2461, sec 8.3]
1889 		 */
1890 		if (code == ND_REDIRECT_ROUTER)
1891 			ln_router = 1;
1892 		else {
1893 			if (is_new)				/* (6-7) */
1894 				ln_router = 0;
1895 		}
1896 		break;
1897 	case ND_ROUTER_SOLICIT:
1898 		/*
1899 		 * is_router flag must always be cleared.
1900 		 */
1901 		ln_router = 0;
1902 		break;
1903 	case ND_ROUTER_ADVERT:
1904 		/*
1905 		 * Mark an entry with lladdr as a router.
1906 		 */
1907 		if ((!is_new && (old_addr || new_addr)) ||	/* (2-5) */
1908 		    (is_new && new_addr)) {			/* (7) */
1909 			ln_router = 1;
1910 		}
1911 		break;
1912 	}
1913 
1914 	return (ln_router);
1915 }
1916 
1917 /*
1918  * Create neighbor cache entry and cache link-layer address,
1919  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1920  *
1921  * type - ICMP6 type
1922  * code - type dependent information
1923  *
1924  */
1925 void
1926 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1927     int lladdrlen, int type, int code)
1928 {
1929 	struct llentry *ln = NULL, *ln_tmp;
1930 	int is_newentry;
1931 	int do_update;
1932 	int olladdr;
1933 	int llchange;
1934 	int flags;
1935 	uint16_t router = 0;
1936 	struct sockaddr_in6 sin6;
1937 	struct mbuf *chain = NULL;
1938 	u_char linkhdr[LLE_MAX_LINKHDR];
1939 	size_t linkhdrsize;
1940 	int lladdr_off;
1941 
1942 	NET_EPOCH_ASSERT();
1943 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1944 
1945 	KASSERT(ifp != NULL, ("%s: ifp == NULL", __func__));
1946 	KASSERT(from != NULL, ("%s: from == NULL", __func__));
1947 
1948 	/* nothing must be updated for unspecified address */
1949 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1950 		return;
1951 
1952 	/*
1953 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1954 	 * the caller.
1955 	 *
1956 	 * XXX If the link does not have link-layer adderss, what should
1957 	 * we do? (ifp->if_addrlen == 0)
1958 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1959 	 * description on it in NS section (RFC 2461 7.2.3).
1960 	 */
1961 	flags = lladdr ? LLE_EXCLUSIVE : 0;
1962 	ln = nd6_lookup(from, flags, ifp);
1963 	is_newentry = 0;
1964 	if (ln == NULL) {
1965 		flags |= LLE_EXCLUSIVE;
1966 		ln = nd6_alloc(from, 0, ifp);
1967 		if (ln == NULL)
1968 			return;
1969 
1970 		/*
1971 		 * Since we already know all the data for the new entry,
1972 		 * fill it before insertion.
1973 		 */
1974 		if (lladdr != NULL) {
1975 			linkhdrsize = sizeof(linkhdr);
1976 			if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
1977 			    linkhdr, &linkhdrsize, &lladdr_off) != 0)
1978 				return;
1979 			lltable_set_entry_addr(ifp, ln, linkhdr, linkhdrsize,
1980 			    lladdr_off);
1981 		}
1982 
1983 		IF_AFDATA_WLOCK(ifp);
1984 		LLE_WLOCK(ln);
1985 		/* Prefer any existing lle over newly-created one */
1986 		ln_tmp = nd6_lookup(from, LLE_EXCLUSIVE, ifp);
1987 		if (ln_tmp == NULL)
1988 			lltable_link_entry(LLTABLE6(ifp), ln);
1989 		IF_AFDATA_WUNLOCK(ifp);
1990 		if (ln_tmp == NULL) {
1991 			/* No existing lle, mark as new entry (6,7) */
1992 			is_newentry = 1;
1993 			if (lladdr != NULL) {	/* (7) */
1994 				nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
1995 				EVENTHANDLER_INVOKE(lle_event, ln,
1996 				    LLENTRY_RESOLVED);
1997 			}
1998 		} else {
1999 			lltable_free_entry(LLTABLE6(ifp), ln);
2000 			ln = ln_tmp;
2001 			ln_tmp = NULL;
2002 		}
2003 	}
2004 	/* do nothing if static ndp is set */
2005 	if ((ln->la_flags & LLE_STATIC)) {
2006 		if (flags & LLE_EXCLUSIVE)
2007 			LLE_WUNLOCK(ln);
2008 		else
2009 			LLE_RUNLOCK(ln);
2010 		return;
2011 	}
2012 
2013 	olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
2014 	if (olladdr && lladdr) {
2015 		llchange = bcmp(lladdr, ln->ll_addr,
2016 		    ifp->if_addrlen);
2017 	} else if (!olladdr && lladdr)
2018 		llchange = 1;
2019 	else
2020 		llchange = 0;
2021 
2022 	/*
2023 	 * newentry olladdr  lladdr  llchange	(*=record)
2024 	 *	0	n	n	--	(1)
2025 	 *	0	y	n	--	(2)
2026 	 *	0	n	y	y	(3) * STALE
2027 	 *	0	y	y	n	(4) *
2028 	 *	0	y	y	y	(5) * STALE
2029 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
2030 	 *	1	--	y	--	(7) * STALE
2031 	 */
2032 
2033 	do_update = 0;
2034 	if (is_newentry == 0 && llchange != 0) {
2035 		do_update = 1;	/* (3,5) */
2036 
2037 		/*
2038 		 * Record source link-layer address
2039 		 * XXX is it dependent to ifp->if_type?
2040 		 */
2041 		linkhdrsize = sizeof(linkhdr);
2042 		if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
2043 		    linkhdr, &linkhdrsize, &lladdr_off) != 0)
2044 			return;
2045 
2046 		if (lltable_try_set_entry_addr(ifp, ln, linkhdr, linkhdrsize,
2047 		    lladdr_off) == 0) {
2048 			/* Entry was deleted */
2049 			return;
2050 		}
2051 
2052 		nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
2053 
2054 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2055 
2056 		if (ln->la_hold != NULL)
2057 			nd6_grab_holdchain(ln, &chain, &sin6);
2058 	}
2059 
2060 	/* Calculates new router status */
2061 	router = nd6_is_router(type, code, is_newentry, olladdr,
2062 	    lladdr != NULL ? 1 : 0, ln->ln_router);
2063 
2064 	ln->ln_router = router;
2065 	/* Mark non-router redirects with special flag */
2066 	if ((type & 0xFF) == ND_REDIRECT && code != ND_REDIRECT_ROUTER)
2067 		ln->la_flags |= LLE_REDIRECT;
2068 
2069 	if (flags & LLE_EXCLUSIVE)
2070 		LLE_WUNLOCK(ln);
2071 	else
2072 		LLE_RUNLOCK(ln);
2073 
2074 	if (chain != NULL)
2075 		nd6_flush_holdchain(ifp, chain, &sin6);
2076 
2077 	/*
2078 	 * When the link-layer address of a router changes, select the
2079 	 * best router again.  In particular, when the neighbor entry is newly
2080 	 * created, it might affect the selection policy.
2081 	 * Question: can we restrict the first condition to the "is_newentry"
2082 	 * case?
2083 	 * XXX: when we hear an RA from a new router with the link-layer
2084 	 * address option, defrouter_select_fib() is called twice, since
2085 	 * defrtrlist_update called the function as well.  However, I believe
2086 	 * we can compromise the overhead, since it only happens the first
2087 	 * time.
2088 	 * XXX: although defrouter_select_fib() should not have a bad effect
2089 	 * for those are not autoconfigured hosts, we explicitly avoid such
2090 	 * cases for safety.
2091 	 */
2092 	if ((do_update || is_newentry) && router &&
2093 	    ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
2094 		/*
2095 		 * guaranteed recursion
2096 		 */
2097 		defrouter_select_fib(ifp->if_fib);
2098 	}
2099 }
2100 
2101 static void
2102 nd6_slowtimo(void *arg)
2103 {
2104 	struct epoch_tracker et;
2105 	CURVNET_SET((struct vnet *) arg);
2106 	struct nd_ifinfo *nd6if;
2107 	struct ifnet *ifp;
2108 
2109 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
2110 	    nd6_slowtimo, curvnet);
2111 	NET_EPOCH_ENTER(et);
2112 	CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2113 		if (ifp->if_afdata[AF_INET6] == NULL)
2114 			continue;
2115 		nd6if = ND_IFINFO(ifp);
2116 		if (nd6if->basereachable && /* already initialized */
2117 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
2118 			/*
2119 			 * Since reachable time rarely changes by router
2120 			 * advertisements, we SHOULD insure that a new random
2121 			 * value gets recomputed at least once every few hours.
2122 			 * (RFC 2461, 6.3.4)
2123 			 */
2124 			nd6if->recalctm = V_nd6_recalc_reachtm_interval;
2125 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
2126 		}
2127 	}
2128 	NET_EPOCH_EXIT(et);
2129 	CURVNET_RESTORE();
2130 }
2131 
2132 void
2133 nd6_grab_holdchain(struct llentry *ln, struct mbuf **chain,
2134     struct sockaddr_in6 *sin6)
2135 {
2136 
2137 	LLE_WLOCK_ASSERT(ln);
2138 
2139 	*chain = ln->la_hold;
2140 	ln->la_hold = NULL;
2141 	lltable_fill_sa_entry(ln, (struct sockaddr *)sin6);
2142 
2143 	if (ln->ln_state == ND6_LLINFO_STALE) {
2144 		/*
2145 		 * The first time we send a packet to a
2146 		 * neighbor whose entry is STALE, we have
2147 		 * to change the state to DELAY and a sets
2148 		 * a timer to expire in DELAY_FIRST_PROBE_TIME
2149 		 * seconds to ensure do neighbor unreachability
2150 		 * detection on expiration.
2151 		 * (RFC 2461 7.3.3)
2152 		 */
2153 		nd6_llinfo_setstate(ln, ND6_LLINFO_DELAY);
2154 	}
2155 }
2156 
2157 int
2158 nd6_output_ifp(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
2159     struct sockaddr_in6 *dst, struct route *ro)
2160 {
2161 	int error;
2162 	int ip6len;
2163 	struct ip6_hdr *ip6;
2164 	struct m_tag *mtag;
2165 
2166 #ifdef MAC
2167 	mac_netinet6_nd6_send(ifp, m);
2168 #endif
2169 
2170 	/*
2171 	 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
2172 	 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
2173 	 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
2174 	 * to be diverted to user space.  When re-injected into the kernel,
2175 	 * send_output() will directly dispatch them to the outgoing interface.
2176 	 */
2177 	if (send_sendso_input_hook != NULL) {
2178 		mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
2179 		if (mtag != NULL) {
2180 			ip6 = mtod(m, struct ip6_hdr *);
2181 			ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
2182 			/* Use the SEND socket */
2183 			error = send_sendso_input_hook(m, ifp, SND_OUT,
2184 			    ip6len);
2185 			/* -1 == no app on SEND socket */
2186 			if (error == 0 || error != -1)
2187 			    return (error);
2188 		}
2189 	}
2190 
2191 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
2192 	IP_PROBE(send, NULL, NULL, mtod(m, struct ip6_hdr *), ifp, NULL,
2193 	    mtod(m, struct ip6_hdr *));
2194 
2195 	if ((ifp->if_flags & IFF_LOOPBACK) == 0)
2196 		origifp = ifp;
2197 
2198 	error = (*ifp->if_output)(origifp, m, (struct sockaddr *)dst, ro);
2199 	return (error);
2200 }
2201 
2202 /*
2203  * Lookup link headerfor @sa_dst address. Stores found
2204  * data in @desten buffer. Copy of lle ln_flags can be also
2205  * saved in @pflags if @pflags is non-NULL.
2206  *
2207  * If destination LLE does not exists or lle state modification
2208  * is required, call "slow" version.
2209  *
2210  * Return values:
2211  * - 0 on success (address copied to buffer).
2212  * - EWOULDBLOCK (no local error, but address is still unresolved)
2213  * - other errors (alloc failure, etc)
2214  */
2215 int
2216 nd6_resolve(struct ifnet *ifp, int is_gw, struct mbuf *m,
2217     const struct sockaddr *sa_dst, u_char *desten, uint32_t *pflags,
2218     struct llentry **plle)
2219 {
2220 	struct llentry *ln = NULL;
2221 	const struct sockaddr_in6 *dst6;
2222 
2223 	NET_EPOCH_ASSERT();
2224 
2225 	if (pflags != NULL)
2226 		*pflags = 0;
2227 
2228 	dst6 = (const struct sockaddr_in6 *)sa_dst;
2229 
2230 	/* discard the packet if IPv6 operation is disabled on the interface */
2231 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
2232 		m_freem(m);
2233 		return (ENETDOWN); /* better error? */
2234 	}
2235 
2236 	if (m != NULL && m->m_flags & M_MCAST) {
2237 		switch (ifp->if_type) {
2238 		case IFT_ETHER:
2239 		case IFT_L2VLAN:
2240 		case IFT_BRIDGE:
2241 			ETHER_MAP_IPV6_MULTICAST(&dst6->sin6_addr,
2242 						 desten);
2243 			return (0);
2244 		default:
2245 			m_freem(m);
2246 			return (EAFNOSUPPORT);
2247 		}
2248 	}
2249 
2250 	ln = nd6_lookup(&dst6->sin6_addr, plle ? LLE_EXCLUSIVE : LLE_UNLOCKED,
2251 	    ifp);
2252 	if (ln != NULL && (ln->r_flags & RLLE_VALID) != 0) {
2253 		/* Entry found, let's copy lle info */
2254 		bcopy(ln->r_linkdata, desten, ln->r_hdrlen);
2255 		if (pflags != NULL)
2256 			*pflags = LLE_VALID | (ln->r_flags & RLLE_IFADDR);
2257 		/* Check if we have feedback request from nd6 timer */
2258 		if (ln->r_skip_req != 0) {
2259 			LLE_REQ_LOCK(ln);
2260 			ln->r_skip_req = 0; /* Notify that entry was used */
2261 			ln->lle_hittime = time_uptime;
2262 			LLE_REQ_UNLOCK(ln);
2263 		}
2264 		if (plle) {
2265 			LLE_ADDREF(ln);
2266 			*plle = ln;
2267 			LLE_WUNLOCK(ln);
2268 		}
2269 		return (0);
2270 	} else if (plle && ln)
2271 		LLE_WUNLOCK(ln);
2272 
2273 	return (nd6_resolve_slow(ifp, 0, m, dst6, desten, pflags, plle));
2274 }
2275 
2276 /*
2277  * Do L2 address resolution for @sa_dst address. Stores found
2278  * address in @desten buffer. Copy of lle ln_flags can be also
2279  * saved in @pflags if @pflags is non-NULL.
2280  *
2281  * Heavy version.
2282  * Function assume that destination LLE does not exist,
2283  * is invalid or stale, so LLE_EXCLUSIVE lock needs to be acquired.
2284  *
2285  * Set noinline to be dtrace-friendly
2286  */
2287 static __noinline int
2288 nd6_resolve_slow(struct ifnet *ifp, int flags, struct mbuf *m,
2289     const struct sockaddr_in6 *dst, u_char *desten, uint32_t *pflags,
2290     struct llentry **plle)
2291 {
2292 	struct llentry *lle = NULL, *lle_tmp;
2293 	struct in6_addr *psrc, src;
2294 	int send_ns, ll_len;
2295 	char *lladdr;
2296 
2297 	NET_EPOCH_ASSERT();
2298 
2299 	/*
2300 	 * Address resolution or Neighbor Unreachability Detection
2301 	 * for the next hop.
2302 	 * At this point, the destination of the packet must be a unicast
2303 	 * or an anycast address(i.e. not a multicast).
2304 	 */
2305 	if (lle == NULL) {
2306 		lle = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp);
2307 		if ((lle == NULL) && nd6_is_addr_neighbor(dst, ifp))  {
2308 			/*
2309 			 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
2310 			 * the condition below is not very efficient.  But we believe
2311 			 * it is tolerable, because this should be a rare case.
2312 			 */
2313 			lle = nd6_alloc(&dst->sin6_addr, 0, ifp);
2314 			if (lle == NULL) {
2315 				char ip6buf[INET6_ADDRSTRLEN];
2316 				log(LOG_DEBUG,
2317 				    "nd6_output: can't allocate llinfo for %s "
2318 				    "(ln=%p)\n",
2319 				    ip6_sprintf(ip6buf, &dst->sin6_addr), lle);
2320 				m_freem(m);
2321 				return (ENOBUFS);
2322 			}
2323 
2324 			IF_AFDATA_WLOCK(ifp);
2325 			LLE_WLOCK(lle);
2326 			/* Prefer any existing entry over newly-created one */
2327 			lle_tmp = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp);
2328 			if (lle_tmp == NULL)
2329 				lltable_link_entry(LLTABLE6(ifp), lle);
2330 			IF_AFDATA_WUNLOCK(ifp);
2331 			if (lle_tmp != NULL) {
2332 				lltable_free_entry(LLTABLE6(ifp), lle);
2333 				lle = lle_tmp;
2334 				lle_tmp = NULL;
2335 			}
2336 		}
2337 	}
2338 	if (lle == NULL) {
2339 		m_freem(m);
2340 		return (ENOBUFS);
2341 	}
2342 
2343 	LLE_WLOCK_ASSERT(lle);
2344 
2345 	/*
2346 	 * The first time we send a packet to a neighbor whose entry is
2347 	 * STALE, we have to change the state to DELAY and a sets a timer to
2348 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2349 	 * neighbor unreachability detection on expiration.
2350 	 * (RFC 2461 7.3.3)
2351 	 */
2352 	if (lle->ln_state == ND6_LLINFO_STALE)
2353 		nd6_llinfo_setstate(lle, ND6_LLINFO_DELAY);
2354 
2355 	/*
2356 	 * If the neighbor cache entry has a state other than INCOMPLETE
2357 	 * (i.e. its link-layer address is already resolved), just
2358 	 * send the packet.
2359 	 */
2360 	if (lle->ln_state > ND6_LLINFO_INCOMPLETE) {
2361 		if (flags & LLE_ADDRONLY) {
2362 			lladdr = lle->ll_addr;
2363 			ll_len = ifp->if_addrlen;
2364 		} else {
2365 			lladdr = lle->r_linkdata;
2366 			ll_len = lle->r_hdrlen;
2367 		}
2368 		bcopy(lladdr, desten, ll_len);
2369 		if (pflags != NULL)
2370 			*pflags = lle->la_flags;
2371 		if (plle) {
2372 			LLE_ADDREF(lle);
2373 			*plle = lle;
2374 		}
2375 		LLE_WUNLOCK(lle);
2376 		return (0);
2377 	}
2378 
2379 	/*
2380 	 * There is a neighbor cache entry, but no ethernet address
2381 	 * response yet.  Append this latest packet to the end of the
2382 	 * packet queue in the mbuf.  When it exceeds nd6_maxqueuelen,
2383 	 * the oldest packet in the queue will be removed.
2384 	 */
2385 
2386 	if (lle->la_hold != NULL) {
2387 		struct mbuf *m_hold;
2388 		int i;
2389 
2390 		i = 0;
2391 		for (m_hold = lle->la_hold; m_hold; m_hold = m_hold->m_nextpkt){
2392 			i++;
2393 			if (m_hold->m_nextpkt == NULL) {
2394 				m_hold->m_nextpkt = m;
2395 				break;
2396 			}
2397 		}
2398 		while (i >= V_nd6_maxqueuelen) {
2399 			m_hold = lle->la_hold;
2400 			lle->la_hold = lle->la_hold->m_nextpkt;
2401 			m_freem(m_hold);
2402 			i--;
2403 		}
2404 	} else {
2405 		lle->la_hold = m;
2406 	}
2407 
2408 	/*
2409 	 * If there has been no NS for the neighbor after entering the
2410 	 * INCOMPLETE state, send the first solicitation.
2411 	 * Note that for newly-created lle la_asked will be 0,
2412 	 * so we will transition from ND6_LLINFO_NOSTATE to
2413 	 * ND6_LLINFO_INCOMPLETE state here.
2414 	 */
2415 	psrc = NULL;
2416 	send_ns = 0;
2417 	if (lle->la_asked == 0) {
2418 		lle->la_asked++;
2419 		send_ns = 1;
2420 		psrc = nd6_llinfo_get_holdsrc(lle, &src);
2421 
2422 		nd6_llinfo_setstate(lle, ND6_LLINFO_INCOMPLETE);
2423 	}
2424 	LLE_WUNLOCK(lle);
2425 	if (send_ns != 0)
2426 		nd6_ns_output(ifp, psrc, NULL, &dst->sin6_addr, NULL);
2427 
2428 	return (EWOULDBLOCK);
2429 }
2430 
2431 /*
2432  * Do L2 address resolution for @sa_dst address. Stores found
2433  * address in @desten buffer. Copy of lle ln_flags can be also
2434  * saved in @pflags if @pflags is non-NULL.
2435  *
2436  * Return values:
2437  * - 0 on success (address copied to buffer).
2438  * - EWOULDBLOCK (no local error, but address is still unresolved)
2439  * - other errors (alloc failure, etc)
2440  */
2441 int
2442 nd6_resolve_addr(struct ifnet *ifp, int flags, const struct sockaddr *dst,
2443     char *desten, uint32_t *pflags)
2444 {
2445 	int error;
2446 
2447 	flags |= LLE_ADDRONLY;
2448 	error = nd6_resolve_slow(ifp, flags, NULL,
2449 	    (const struct sockaddr_in6 *)dst, desten, pflags, NULL);
2450 	return (error);
2451 }
2452 
2453 int
2454 nd6_flush_holdchain(struct ifnet *ifp, struct mbuf *chain,
2455     struct sockaddr_in6 *dst)
2456 {
2457 	struct mbuf *m, *m_head;
2458 	int error = 0;
2459 
2460 	m_head = chain;
2461 
2462 	while (m_head) {
2463 		m = m_head;
2464 		m_head = m_head->m_nextpkt;
2465 		m->m_nextpkt = NULL;
2466 		error = nd6_output_ifp(ifp, ifp, m, dst, NULL);
2467 	}
2468 
2469 	/*
2470 	 * XXX
2471 	 * note that intermediate errors are blindly ignored
2472 	 */
2473 	return (error);
2474 }
2475 
2476 static int
2477 nd6_need_cache(struct ifnet *ifp)
2478 {
2479 	/*
2480 	 * XXX: we currently do not make neighbor cache on any interface
2481 	 * other than Ethernet and GIF.
2482 	 *
2483 	 * RFC2893 says:
2484 	 * - unidirectional tunnels needs no ND
2485 	 */
2486 	switch (ifp->if_type) {
2487 	case IFT_ETHER:
2488 	case IFT_IEEE1394:
2489 	case IFT_L2VLAN:
2490 	case IFT_INFINIBAND:
2491 	case IFT_BRIDGE:
2492 	case IFT_PROPVIRTUAL:
2493 		return (1);
2494 	default:
2495 		return (0);
2496 	}
2497 }
2498 
2499 /*
2500  * Add pernament ND6 link-layer record for given
2501  * interface address.
2502  *
2503  * Very similar to IPv4 arp_ifinit(), but:
2504  * 1) IPv6 DAD is performed in different place
2505  * 2) It is called by IPv6 protocol stack in contrast to
2506  * arp_ifinit() which is typically called in SIOCSIFADDR
2507  * driver ioctl handler.
2508  *
2509  */
2510 int
2511 nd6_add_ifa_lle(struct in6_ifaddr *ia)
2512 {
2513 	struct ifnet *ifp;
2514 	struct llentry *ln, *ln_tmp;
2515 	struct sockaddr *dst;
2516 
2517 	ifp = ia->ia_ifa.ifa_ifp;
2518 	if (nd6_need_cache(ifp) == 0)
2519 		return (0);
2520 
2521 	dst = (struct sockaddr *)&ia->ia_addr;
2522 	ln = lltable_alloc_entry(LLTABLE6(ifp), LLE_IFADDR, dst);
2523 	if (ln == NULL)
2524 		return (ENOBUFS);
2525 
2526 	IF_AFDATA_WLOCK(ifp);
2527 	LLE_WLOCK(ln);
2528 	/* Unlink any entry if exists */
2529 	ln_tmp = lla_lookup(LLTABLE6(ifp), LLE_EXCLUSIVE, dst);
2530 	if (ln_tmp != NULL)
2531 		lltable_unlink_entry(LLTABLE6(ifp), ln_tmp);
2532 	lltable_link_entry(LLTABLE6(ifp), ln);
2533 	IF_AFDATA_WUNLOCK(ifp);
2534 
2535 	if (ln_tmp != NULL)
2536 		EVENTHANDLER_INVOKE(lle_event, ln_tmp, LLENTRY_EXPIRED);
2537 	EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2538 
2539 	LLE_WUNLOCK(ln);
2540 	if (ln_tmp != NULL)
2541 		llentry_free(ln_tmp);
2542 
2543 	return (0);
2544 }
2545 
2546 /*
2547  * Removes either all lle entries for given @ia, or lle
2548  * corresponding to @ia address.
2549  */
2550 void
2551 nd6_rem_ifa_lle(struct in6_ifaddr *ia, int all)
2552 {
2553 	struct sockaddr_in6 mask, addr;
2554 	struct sockaddr *saddr, *smask;
2555 	struct ifnet *ifp;
2556 
2557 	ifp = ia->ia_ifa.ifa_ifp;
2558 	memcpy(&addr, &ia->ia_addr, sizeof(ia->ia_addr));
2559 	memcpy(&mask, &ia->ia_prefixmask, sizeof(ia->ia_prefixmask));
2560 	saddr = (struct sockaddr *)&addr;
2561 	smask = (struct sockaddr *)&mask;
2562 
2563 	if (all != 0)
2564 		lltable_prefix_free(AF_INET6, saddr, smask, LLE_STATIC);
2565 	else
2566 		lltable_delete_addr(LLTABLE6(ifp), LLE_IFADDR, saddr);
2567 }
2568 
2569 static void
2570 clear_llinfo_pqueue(struct llentry *ln)
2571 {
2572 	struct mbuf *m_hold, *m_hold_next;
2573 
2574 	for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) {
2575 		m_hold_next = m_hold->m_nextpkt;
2576 		m_freem(m_hold);
2577 	}
2578 
2579 	ln->la_hold = NULL;
2580 }
2581 
2582 static int
2583 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2584 {
2585 	struct in6_prefix p;
2586 	struct sockaddr_in6 s6;
2587 	struct nd_prefix *pr;
2588 	struct nd_pfxrouter *pfr;
2589 	time_t maxexpire;
2590 	int error;
2591 	char ip6buf[INET6_ADDRSTRLEN];
2592 
2593 	if (req->newptr)
2594 		return (EPERM);
2595 
2596 	error = sysctl_wire_old_buffer(req, 0);
2597 	if (error != 0)
2598 		return (error);
2599 
2600 	bzero(&p, sizeof(p));
2601 	p.origin = PR_ORIG_RA;
2602 	bzero(&s6, sizeof(s6));
2603 	s6.sin6_family = AF_INET6;
2604 	s6.sin6_len = sizeof(s6);
2605 
2606 	ND6_RLOCK();
2607 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
2608 		p.prefix = pr->ndpr_prefix;
2609 		if (sa6_recoverscope(&p.prefix)) {
2610 			log(LOG_ERR, "scope error in prefix list (%s)\n",
2611 			    ip6_sprintf(ip6buf, &p.prefix.sin6_addr));
2612 			/* XXX: press on... */
2613 		}
2614 		p.raflags = pr->ndpr_raf;
2615 		p.prefixlen = pr->ndpr_plen;
2616 		p.vltime = pr->ndpr_vltime;
2617 		p.pltime = pr->ndpr_pltime;
2618 		p.if_index = pr->ndpr_ifp->if_index;
2619 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2620 			p.expire = 0;
2621 		else {
2622 			/* XXX: we assume time_t is signed. */
2623 			maxexpire = (-1) &
2624 			    ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1));
2625 			if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate)
2626 				p.expire = pr->ndpr_lastupdate +
2627 				    pr->ndpr_vltime +
2628 				    (time_second - time_uptime);
2629 			else
2630 				p.expire = maxexpire;
2631 		}
2632 		p.refcnt = pr->ndpr_addrcnt;
2633 		p.flags = pr->ndpr_stateflags;
2634 		p.advrtrs = 0;
2635 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry)
2636 			p.advrtrs++;
2637 		error = SYSCTL_OUT(req, &p, sizeof(p));
2638 		if (error != 0)
2639 			break;
2640 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2641 			s6.sin6_addr = pfr->router->rtaddr;
2642 			if (sa6_recoverscope(&s6))
2643 				log(LOG_ERR,
2644 				    "scope error in prefix list (%s)\n",
2645 				    ip6_sprintf(ip6buf, &pfr->router->rtaddr));
2646 			error = SYSCTL_OUT(req, &s6, sizeof(s6));
2647 			if (error != 0)
2648 				goto out;
2649 		}
2650 	}
2651 out:
2652 	ND6_RUNLOCK();
2653 	return (error);
2654 }
2655 SYSCTL_PROC(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2656 	CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
2657 	NULL, 0, nd6_sysctl_prlist, "S,in6_prefix",
2658 	"NDP prefix list");
2659 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2660 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2661 SYSCTL_INT(_net_inet6_icmp6, OID_AUTO, nd6_gctimer,
2662 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_gctimer), (60 * 60 * 24), "");
2663